Clusterwise Pricing in Stores of a Retail Chain

Abstract

Clusterwise pricing is characterized by prices of each brand of a category being equal in stores belonging to the same cluster. Expected sales necessary to compute profits are estimated using coefficients of a multilayer perceptron which performs better than several parametric models. Store-specific coefficients of sales response models are estimated by a MCMC method. Both assignment of stores to clusters and prices of each cluster are determined by means of improving hit-and-run, a stochastic optimization method. The objective function to be optimized includes both profits and adherence to the usual price level at individual stores. Based on empirical data on sales, prices and marginal costs of a retail chain it is demonstrated that for a moderate level of risk aversion clusterwise pricing leads to higher expected utility than micromarketing pricing with different prices of each brand across individual stores. Clusterwise pricing attains a high percentage of the profits generated by micromarketing pricing and entails lower menu costs than micromarketing pricing.